Enhancement of VSELP Coded Speech under Background Noise

The qrcalily of reproduced speech by VSELP is degraded when it is coded under background noise. We analyzed this phenomena and found that the degradation was caused by the fIuctuation of LPC synthesis filter characteristics. To solve the problem, we introduce spectral snoothing while the input is detected as only mise (silence with background noise). This simple nick successjidly suppressed the artifacts and improved the subjective quality of the reproduced speech.